Blogmark

I enjoyed reading about this algorithm because it demonstrates some clever tricks and problem solving. It's useful to read about new algorithms from time to time if for no other reason than as a source of inspiration to keep you actively thinking about unique was to solve problems.

A high level summary of what makes it so neat: the algorithm uses adaptive bias adjustment, variable-length encoding, and delta compression to achieve extremely dense (in the information-theoretic sense) results for both single-script domains (like German bücher-café.de or Chinese 北京-旅游.cn) and mixed-script domains (like hello世界.com). What makes it particularly elegant is that it does this without any shared state between encoder and decoder. The adaptation is purely a deterministic function of the encoded data itself. This is a really cool example of how you can use the constraints of a system to your advantage to solve a problem.